Skip links

Give your AI assistant direct access to official company registries.

Connect Zephira to Claude, Cursor, or any MCP client and ask questions in plain language. Answers come back as structured records from 400+ government registries — with the filing date and source attached to every field.

zephira-mcp / connected MCP · HTTP
Analyst
Tool call
verify_entity Read only
{ "name": "Meridian Freight Inc.",
  "checks": ["existence", "vat", "tax_id"] }
existenceconfirmed
statusActive · Delaware SOS
vat_idvalid · VIES
ein84-3921776
Answer
The entity is real and active in Delaware. Its VAT registration validates, and the EIN comes back from the filed record.
Verified against Delaware SOS & VIES · 24 Jul 2026

Registry data trusted by

LSEG Experian Allianz PwC Kroll Schneider Electric Cognizant Amadeus Retarus
Why an MCP server

Your assistant stops guessing about companies

General-purpose models answer company questions from whatever they absorbed during training. That's stale by definition and impossible to audit. The MCP server replaces recall with a lookup.

Without the server

"I believe the registered office is in Amsterdam, but you should verify this." No registry number. No filing date. No way to show a regulator where the answer came from.

With the server

Every field carries its registry of origin, the official identifier, and the date it was filed. The assistant cites the source because the source came back with the data.

Refreshed against the registry, not a cache

Filings, officer changes, status flags and financial statements land as the registry publishes them. Subscribe to a company and the assistant can tell you what moved.

200+ countries in one interface

One tool call, whatever the jurisdiction. No per-country portals, no bilingual filing formats, no manual normalisation of officer roles or entity types.

16 tools

What your assistant can call

Each tool returns structured JSON with the source registry, official identifier and filing date included in the response.

Search, identity & filings

verify_entity

One call to confirm an entity exists, check its status, and validate its VAT and tax identifiers.

search_companies

Find entities by name, registration number, VAT, LEI or ticker across every covered jurisdiction.

get_company_profile

Full statutory record: legal name, status, incorporation date, registered address, entity type.

resolve_entity

Match a messy internal string to a single registry record, with a confidence score and rejected candidates.

lookup_vat_number

Validate a VAT identifier and return the registered name and address behind it.

lookup_lei

Resolve a Legal Entity Identifier and its registration status.

get_financials

Filed accounts — revenue, assets, equity, employees — normalised across reporting standards.

get_filing_history

Chronological list of documents lodged with the registry, with dates and types.

Ownership & control

get_shareholders

Registered shareholders with holding percentages and share classes as filed.

get_officers

Directors, secretaries and authorised signatories with appointment and resignation dates.

get_officer_profile

Every directorship held by one person across jurisdictions, current and historic.

get_group_structure

Parents and subsidiaries assembled into a single ownership tree.

Monitoring & bulk work

subscribe_to_company

Watch an entity and receive changes to officers, ownership, status or accounts.

list_subscriptions

Review everything currently monitored and when each was last checked.

get_recent_changes

What moved across your watchlist in a given window, grouped by change type.

batch_enrich

Push a list of company names or numbers and get matched registry records back.

Setup

Connected in about two minutes

Remote server, hosted by Zephira. Nothing to run locally, nothing to update.

1

Get an API key

Create a Zephira account and generate a key from Settings → API. The free tier covers evaluation.

2

Add the server

Paste the config below into your MCP client, or add the remote URL through your client's connector settings.

3

Ask a question

Restart the client and ask about any company. The assistant picks the right tool on its own.

// claude_desktop_config.json
{
  "mcpServers": {
    "zephira": {
      "url": "https://mcp.zephira.ai/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}
# Claude Code
claude mcp add --transport http zephira https://mcp.zephira.ai/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"
In practice

What teams actually ask it

"Onboard this supplier list — flag anything dissolved, sanctioned or under 12 months old."

KYB onboarding

Batch enrichment against the registry of record, with status and incorporation date checked in the same pass.

"Trace ownership of this Cyprus entity up to natural persons and tell me where it breaks."

UBO investigation

Layer-by-layer traversal that names the jurisdiction where disclosure stops, instead of returning an empty result.

"Anything change across my 4,000 monitored counterparties this week?"

Portfolio monitoring

Registry events summarised by materiality, so review time goes to the twelve that moved.

"Build me every active manufacturer in Poland with 50–250 employees and filed accounts."

Market sizing

Statutory filters rather than self-reported profiles — the population is the registry population.

"Do these three vendors share a director or a registered address?"

Fraud & collusion checks

Connection analysis across officers, addresses and shareholders, in a single question.

"Pull filed revenue and headcount for these 40 targets, last three years."

Diligence prep

Filed accounts normalised across reporting standards, delivered into the document you're already writing.

Governance

Built for teams that get audited

An assistant with data access needs the same controls as any other system touching regulated workflows.

  • Read-only by design. No tool in the server can write to or modify your systems.
  • Provenance on every field. Source registry, official identifier and filing date travel with the record.
  • Full call logging. Every tool invocation is recorded against your key for audit and cost review.
  • Your prompts aren't training data. Queries are used to serve the request and nothing else.
  • Scoped keys. Issue separate keys per team or environment and revoke any of them independently.
  • Licensed for the use you have. Internal, redistribution and AI-grounding rights are handled in contract, not left ambiguous.
FAQ

Questions worth asking first

What is MCP, in one sentence?

The Model Context Protocol is an open standard that lets an AI assistant call external tools and data sources directly. Zephira runs a server that speaks it, so any compatible assistant can query official registry data the same way it would call any other tool.

Which clients work with this?

Claude Desktop, Claude Code, Cursor, VS Code with an MCP extension, and anything else that supports remote MCP servers over HTTP. If your client can add a URL and a bearer token, it works.

Where does the data come from?

Primary company registries — 400+ government sources across 200+ countries, covering more than 600 million entities. Records are sourced from the registry that legally holds them, which is why each response carries its own provenance.

How is usage priced?

By tool call, drawn from your account balance, on the same rate card as the API. The free tier is there so you can evaluate before committing. Every call appears in your usage log with the tool name attached.

Can the assistant change anything in my systems?

No. The server exposes read operations only. It retrieves records; it cannot write, delete or push data anywhere.

What happens when a registry doesn't publish something?

The tool returns an explicit gap rather than an estimate. If beneficial ownership isn't disclosed in a jurisdiction, the response says so and names the layer where the chain stops.

How is this different from letting the assistant search the web?

Web results give you a page about a company. The server gives you the company's statutory record — typed fields, the official identifier, and the date the filing was lodged. One is something to read and verify; the other is something you can put in a compliance file.

Do I need this if we already use the Zephira API?

They're the same data behind two different front doors. The API is for systems you build. The MCP server is for the analyst who wants an answer inside the tool they're already writing in, without waiting on an engineering ticket. Most teams end up running both.

How do I control cost and rate limits?

Calls draw from your account balance and appear individually in the usage log with the tool name attached. Set a monthly ceiling per key, and issue separate keys per team so you can see where consumption is going before you scale it.

How current is the data?

Refresh cadence follows the registry. Some publish changes daily, others in weekly or monthly batches, and a few only on request. Every record returns the date it was filed and the date it was last checked, so you can see the age of an answer rather than assume it.

These records name individuals. How does that sit with GDPR?

Officer and shareholder details come from statutory registers that are published by the registry itself. Processing is on that legal footing, not on consent. Your own downstream use still needs its own basis — the licence terms set out what you can do with the records.

Can we use the output to ground our own AI product?

Grounding, redistribution and model training are three separate rights and they're priced separately. Internal use is covered by the standard licence. Anything customer-facing needs the right terms in place first — the commercial team will scope it rather than leave it implied.

What if a company name matches several entities?

resolve_entity returns the best match with a confidence score, plus the candidates it rejected and why. The assistant surfaces the ambiguity instead of silently picking one, which is the failure mode that matters in onboarding.

How far back do the records go?

Depth varies by registry. Most major jurisdictions carry filing history, officer appointments and resignations, and previous names going back years; several carry the full lifetime of the entity. The coverage documentation sets out what's available per jurisdiction before you build a process on it.

How does this compare with the incumbents?

The large providers sell an interpretation layer — scores, matched identifiers, proprietary hierarchies — built on top of source records. That's valuable, and it's also the thing you can't show a regulator. This server returns the underlying registry record with its provenance intact. Different job, and often a complementary one.

Can it run inside our own environment?

The hosted remote server covers most teams. Where policy requires it, a self-hosted deployment against your own key is available on enterprise agreements — talk to the team about what your controls actually demand.

What happens if the server is unavailable?

Tool calls fail cleanly with an error the assistant can report, rather than returning a partial or stale record. Status and incident history are published, and enterprise agreements carry an availability commitment.

Ask your first company question today

Create a key, paste the config, and let your assistant pull statutory records instead of recalling them.

This website uses cookies to improve your web experience.